Monte Carlo forecasts from a VECM

[27npg]

I am looking to estimate a Monte Carlo forecasts from a Vector Error Correction Model (VECM) over say a 5 year period. I was wondering if theres a way of doing this in EViews either through a pre-programmed function or in terms of creating a new code?

Any help would be greatly appreciated.

NPG

[EViews Gareth]

What do you want to Monte Carlo? (i.e. what do you want to randomize, and how?)

[27npg]

Hi Gareth - thanks for your reply.

I am basically trying to generate a Monte Carlo forecasts from the estimated VEC model over a 5-year horizon. Ultimately, I am trying to get a 5-by-3-by-1000 numeric array of simulated values of the response series in which the rows correspond to periods in the forecast horizon (5 years), the columns correspond to the series in Y, and pages correspond to simulated paths.

Hope this makes sense!

NPG

[EViews Gareth]

It does, but what do you want to draw the random numbers on, and from which distribution etc...

[27npg]

Hi Gareth - Gaussian distribution. Drawn from the output responses Y from a conventional, unconditional Monte Carlo simulation

[EViews Gareth]

From your VEC click on Proc->Make Model.

In the model click on Solve to bring up the solve dialog.

On the basic options page, change the "simulation type" to stochastic. Change the "Solution Sample" to be the forecast dates you want.
On the "Stoch. Options" tab of the solve dialog type "myMC" into the "Page name" field.

Hit OK, then look at the new page that was created. Each series is the stacked version of the simulations.

[27npg]

Thats great Gareth - Thanks so much for your help. Much appreciated.

[creagan52]

Just as a quick follow up to this question. I followed these steps and it worked perfertly - thanks you both.

I am however interested in looking at the extreme quantiles of the simualtions. I am able to view these through Graph->Panel Options (clicking median plus quantiles) etc. Is there anyway of getting the actual data for these graphs as a time series as I am looking to plot these onto the original data to show the different paths (similar to what is done in the attached screenshot for another software package).

For now - I can only see the stacked data (1,000s of observations)

[EViews Gareth]

If you open one of the forecast series in the panel page you can click on View->Stats by classification and then use @obsid or @date as the classifier in order to view statistics (means, median, quantiles) across the simulations.

You can also use the by-group functions to generate new time series with the same stats.
http://www.eviews.com/help/helpintro.ht ... cs.html%23