# Stationary vs. Non-stationary data in a BVAR model

I am replicating a paper using BVAR model and I first I have run the model with non-stationary data. Then I just wanted to compare the results with stationary data and I launched the model but I get this error message:

model1s <- uhlig.reject(Y=data2, nlags=12, draws=200, subdraws=200, nkeep=1000, KMIN=1,

•                     KMAX=5, constrained=constr, constant=FALSE, steps=60) Starting MCMC, Sun Jul 25 20:30:22 2021.


|======================================================================================================| 100% Error in uhlig.reject(Y = data2, nlags = 12, draws = 200, subdraws = 200, : Not enough accepted draws to proceed! In addition: There were 50 or more warnings (use warnings() to see the first 50)

I am still very new into Bayesian statistic and I wonder why I cannot reproduce the result with stationary data? Loosely speaking this method estimates the effects from monetary policy shock by setting initial constrains on the variables and then for each draw the impulse responses are calculated and if they do not satisfy the constrain the draw is not kept. Note: I do not include data, since it is very long and I think my question is more related to the theory. Thanks!

## 1 Answer

I don't think that the issue is stationarity. In the original paper Uhlig uses logs of integrated variables like the CPI without transforming them further or including a time trend.

From looking at the source code of the function you call here I see that it's running an MCMC, so your error may mean, that the current number of draws you set is too small.

Specifically, if you look at line 135 that throws the error, you see that you can get there, only if no draw has been accepted.

You would very much help yourself and everybody else, if you provide a description of the data and the model that you're trying to fit.