Bayesian impulse responses

Estimation DSGE-VAR
1

Dear Sirs,

I am replicating SW 2003 for the euroarea but I m not able neither to get Bayesian IRf nor finding them on the stored output after:
estimation(datafile=rawdata_euromodel_1, first_obs=5, lik_init=2, mode_compute=0,mode_file=sw_euro_mode,mh_replic=0,dsge_var=0.35, bayesian_irf). Why?

I got IR only if I set stoch_simul(irf=20) after estimation, but can those be considered as Bayesian IRFs?

2

Because you set
mh_replic=0
If there are no MCMC draws, you cannot draw from the posterior for generating Bayesian IRFs

3

Thank you Sir,

As far as a dsge-var version of SW model, since I have data on 7 observables I drop 3 of the 10 shocks and I end up with the covariance matrix of innovations being not p.d. Is there any way of solving this problem apart from loading data on other variables?

4

At which stage do you get this error message?

5

First it tells that Matrix is close to be singular or badly scaled. However the red error message starts later telling:
Error using print_info (line 127)…the matrix of the VAR innovations, based on the artificial sample, is not p.d.

6

This usually implies that not all of your shocks independently affect one of the variables. So you either must use different shocks or different observables.

7

However, the DSGE VAR for the SW with US data was fine, I estimated the dsge prior weight, I got IRs but I didn’ t get the Variance shock decomposition.
The error message is:
Error using dseries/subsref (line 220)
dseries::subsref: Indices are out of bounds! Subsample cannot end after 230Y.
Error in makedataset (line 252)
DynareDataset = DynareDataset(FIRSTOBS:lastobs);…

Is there any way of getting the Variance decomposition of the shocks?

8

That error message indicates that you did not correctly specify the nobs and first_obs-options.

9

1)I have 230 obs (time series of dy), then I specified first_obs=5 and nobs=221 and it works, but I cannot still understand why it does not by setting nobs=222 and greater values.

  1. Moreover, the graph of shock decomposition of y looks strange because the line of the series of the endogeneous vbl y is shifted downward of about 5 points as compared to the same line.
10
  1. Do you have missing values in the data? If not, I might need to see the codes
  2. That usually happens if a constant term is not correctly handled in the observation equations.