This project is read-only.

Median impulse response function

Dec 24, 2014 at 4:35 PM
Hi Jaromir,
I am trying to employ the sign restrictions approach in IRIS to estimate the monetary svar with sign restrictions.
I noticed that the median impulse response function that I get change substantially each time I run the model (I impose the option to compute and plot only the median response).
Could you please explain me why this is happening, and how I can fix that.
Dec 24, 2014 at 4:40 PM
How many successful draws do you get? It could just be that the sign restrictions you impose do not really reflect what's in the data so that the distribution you get is mostly influenced by a chance. Not really much to do about it (haha, I'm not really a fan of any kind of structural VARs, waste of time in my opinion :))

Dec 24, 2014 at 5:25 PM
Sign restrictions are fairly standard: positive shock in interest rate should have positive contemporary effect on interest rate, and negative on output, prices, M1. I run the model about 10 times. Each time I get the result and each time the median IRF
has different shape. From: jaromirbenes
Dec 24, 2014 at 7:10 PM
I didn't say people don't use them :) I just expressed my doubts about the usefulness of the method to learn about the real world :)) (this is of course no offense meant).

Could you post the codes? I'll have a look at your particular application. I might spot a way how to deal with the problem.

Dec 24, 2014 at 7:34 PM
What I mean for instance in this particular case is the following. Most of the changes in policy interest rates observed in the real world are not interest rate shocks but rather systematic movements in response to macroeconomic developments -- e.g. when inflation pressures cumulate central banks raise the rates (think of a policy reaction function). Interest rate shocks in the sense you define them through the sign restrictions account for a minuscule percentage of the observed variations in the macro variables (depending obviously on the country and the historical episode etc). This in turn means that it is very difficult to identify such shocks (you would need very very long time series) and that their estimates will be rather unrobust.

Dec 25, 2014 at 6:56 AM
Merry Christmas. I will post it tomorrow. My wife will kill me if I do that today. From: jaromirbenes
Dec 25, 2014 at 12:28 PM
Haha. My wife almost divorced me for responding to the issue :)

Dec 26, 2014 at 7:18 AM
Modified code file, VAR_sign_restrict, which I used to estimate my model is in attachment together with my data. I am new in this field and I have never used matlab before. So, I have not figured out how to upload my own data file. Hence, I used your code
file, make_up_data, to create 7 time series of artificial data (instead of 3 in your original file) and just copy-paste my data for Lithuania into that file. I started with used the Lithuania data and applied only identification for interest rate shock (positive
contemporary response of interest rate, negative of output, prices and money) to see what I will get. In general, I get results which look as conventional IRF for prices (lcpi), but as I explained above the shape of obtained IRF does change each time I run
the model. From: jaromirbenes
Dec 29, 2014 at 7:04 PM
Please email me your files to jaromir (dot) benes (at) gmail (dot) com.

Jan 20, 2015 at 8:08 AM
Hi Jaromir,
I sent you my files three weeks ago. Have you got them? Is it possible to do anything about this problem?
By the way, in the mean time I checked the results of your original codes more closely and same problem appears there, although changes in IRF are smaller.
Jan 20, 2015 at 7:58 PM
Apologies... This has been and still is a very busy time. I cannot promise anything before mid of February... Really sorry about that. I will look into the issue then...