Dear Aboubkr Med,

I wrote my doctoral thesis about political monetary cycles in Turkey. And my aim was how is change and response monetary indicators ( M0, M1 and etc) totally election periods in Turkey. I used ARIMA or Box-Jenkins Model. It is equal (same method) Response -Impulse analysis. Before and after election shock on monetary indicators I investigated this model. Maybe you can try.

Best regards.

Hi, I will try to elaborate on the previous two answers that seem to be perfect to me.

If you want to analyze the dynamic response of a variable to a given shock, the usual way to go is to implement an impulse-response analysis. The "impulse" is your exogenous shock. The "response" is the dynamic "effect" on the variable of your interest. In stationary environments, the "response" should decay to zero as times goes by. Accordingly, some researchers report the maximum effect, or the cummulative efect. Some other researchers are concerned with the short time effect. Others with longer term effects and you name it. Given that the response is suppose to die out, you can analyze the "duration" of the effect as the "last period for which the effect was different from zero at the 5% significance level", for instance. This of course requires computation of the impulse-response confidence bands, that are readily available in many packages.

Impulse-Response analysis is not resricted to VARs. It can be implemented in general systems, including ARIMA models, DSGEs, and multiple equations models. Furthermore, the "impulse" or initial shock can be modeled as you wish. It can be defined only on one variable, or on multiple variables at the same time. This last "shock is a "multivariate" shock. Moreover, sometimes it is interesting to analyze what happens when you are faced with three shocks in a row of the same magnitude. So, in summary, you need to focus on the implementation of impulse-response analysis.

The case of contemporaneous effects is trickier. One way to go is to model the problem using a Structural VAR as suggested by Tihana.

You can also model "contemporaneous " effects with a traditional VAR but using shocks that are contemporaneous. One direct way to do this is to use the orthogonalized impulse-response analysis that relies on the structure of the correlation matriz of the error terms to designed the type of multivariate shock you will be using.

Interesting discussion, Hope it helps!!

Best Regards!!!

Thank you all for the great suggestions

I have read some articles and textbooks on SVARs, it has many applications, as I understood from the textbooks Blanchard and Quah (1989) application of SVAR has some restriction on the matrix in the short run, based on their example they assumed that GDP shows response to shocks in fiscal policies in the long run. Therefore they set the parameter of this specific relationship to zero. In my case, however, I believe there might be strong response in the short run, and this response should become weaker in the long run, the main point is whether this response completely vanish or only get weaker in the long run, so is this mean that I just have to run simple VAR with no restrictions?

In addition, I also believe that the response to shock is subjected to time, what I mean by that is the change in the market volatility over time can have significant effect on the response, high volatility can exacerbate the response, while low volatility would shorten the time framework of the shock, I'm thinking about developing sub-samples relates to different periods of volatility. But I don't know if there are other ways to deal with this problem instead of making sub-samples.

After doing some research, I've found other application of VAR models which allow for the estimated parameters to change overtime, it's called "Time varying VAR"

As said earlier, I'm already assuming that the response of variable Y to the shock in variable X is dynamic and change over time, so I'm wondering if the time varying VAR could be good solution for this problem.

Hi, yes. Time varying VARs is one to go. Yo can also explore breaks in the relationship. Bai and Perron derived tests to explore múltiple breaks in unknown dates. This strategy will give you support to explore your relationship in different subsamples avoiding the critique of "cherry picking" the subsamples that might be "convenient". Therefore, it provides objectivity to your results.

Very interesting discussion!!!!

Thanks for the great suggestions Mr Pincheira, I will definitely take it into consideration

I would definitely look at the SVAR, structural VAR. wich is more flexible than traditional VAR

Your time series synchronous or non-synchronous?

For non-synchronous time series both VAR and SVAR models are not suitable.

Article Contemporaneous Effects of Non-Synchronous Time Series: VAR ...