Historical decmposition with menu driven commands

[econometricz]

I want to get historical composition graphs for my series to show the impact of an external event on the series. I am new to RATS and was trying to get graphs using menu driven commands in RATS. I am getting forecasted and actual series but not the graphs. How would I do it? I tried using the code example from the manual too. Even that is not working. I would greatly appreciate if anyone can help me with qn.

Thanks!!!

[TomDoan]

That's not something that really lends itself to a menu-driven command. The HISTORY instruction itself is fairly simple; it's really a question of organizing the graphs. Can you show what you did that didn't work?

[econometricz]

Thanks for your reply Tom. I need to get historical decomposition graphs for eight price series due to an external event (animal disease outbreak). I need to get graphs that show the actual values including the event and the forecasted price before the event. I have attached the reference paper that I am using. I need the graphs on the same lines.

I would greatly appreciate if you could help me.

[TomDoan]

That doesn't seem to be doing a full historical decomposition. Instead it's doing the forecast based upon the data before the BSE was discovered (which can be doing using a FORECAST instruction), and comparing that with the actuals. The historical decomposition decomposes that gap into the parts due to each shock component, but since there isn't any good way to isolate a "BSE" component, all the shocks are combined.

[econometricz]

Tom, Thanks for the clarification. That makes things clearer. Could you please tell me how could I use the Forecast option to get such graphs.
Also, when I do the historical decomposition using menu driven commands, I get the actual and forecasted values for the series. Is it the same? Can I plot those values for historical decomposition graphs? How can I be sure that the forecasted values do not include the shock already?

To be specific, I have monthly data from Jan 1990 - Dec 2009 and I have eight price series. I need to analyze the impact of an event of Dec 2003 on those eight price series. Can I do such an analysis using the menu driven commands in RATS?

I appreciate your help.

Thanks!

[moderator]

Tom, Thanks for the clarification. That makes things clearer. Could you please tell me how could I use the Forecast option to get such graphs.

You'll really need to look at some of the relevant sections of the User's Guide and Reference Manual and put the pieces together. For the basic task of computing forecasts and graphing these against the actual values (i.e. the original series), it's just a matter of understanding how the FORECAST and GRAPH instructions work. That's all you need to produce the basic results. Start simple, then you can add options to the GRAPH instructions to customize the look of the graphs, and possibly use SPGRAPH instructions to combine multiple graphs on a page. See the details on the HISTORY instruction to determine if you want to also graph results produced by HISTORY.

How can I be sure that the forecasted values do not include the shock already?

You may need to read Tom Doan's comment again, as the answer should be pretty clear from what he describes.

To elaborate, say the BSE shock starts at some period T+1. It appears that all they are doing is computing dynamic forecasts starting at period "T" (i.e one period prior to the shock). That is, they are computing forecasts for periods T+1, T+2, etc. based only on what is known through period T. Because the shock was not part of the "known" information as of entry T, the forecasts are effectively a "what if the shock hadn't occurred" prediction. So replicating that approach is simply a matter of computing your dynamic forecasts starting at a period prior to the shock (presumably they are only estimating the model using data prior to that point as well).

[econometricz]

Thanks a lot for the explanation. It made things easier for me.

[econometricz]

Tom,

I had one more question.

If I use directed acyclic graphs how should I proceed with historical decomposition? I mean how do I account for the covariance matrix which summarizes the contemporaneous causal patterns between orthogonal innovations ? This is the methodology used in the paper that I had attached.

Thanks a lot for your help!

[TomDoan]

The directed acyclic graphs is a completely separate part of the analysis. It's a non-parametric analysis of causation that's not used in the VAR.

[econometricz]

Hello Tom,

Thanks for your reply. But the papers I am using as reference state this:

"Historical decompositions based on directed graphs and causal patterns decompose the price series into historical shocks in each series of the structural VAR model to determine price responses in a neighborhood (time interval) of the E. coli, BSE, and FMD events"

Could you please help me in understanding this. I appreciate your time and help.

Regards!

[TomDoan]

You would have to check with the authors. In the paper you posted, the historical decomposition is doing what we described above.