EViews.com

How can I combine a univariate & conditional model for static and dynamic forecast and generate its MSFE ?

Could you explain what you mean a bit more thoroughly?

Sure...

I do have a univariate Model: (dlog(x)) c @seas(1) @seas(2) @seas(3) @trend

and a conditional model: (dlog(x)) c dlog(x(-1)) dlog(y) dlog(y(-1)) dlog(z) dlog(z(-1)) @seas(1) @seas(2) @seas(3) @trend

Now I would like to combine these models in bivariate way for S and D forecast and compare the different output and MSFE

Sorry, still don't understand.

How do you want to combine them?

Thx for your fast answers & sorry for the confusion.

I want to evaluate the S&D forecast of each model.

e.g.

FORECAST A: Model1 static forecast vs Model2 static forecast

FORECAST B: Model1 dynamic forecast vs Model2 dynamic forecast

Now, to compare the acuracy I want to focus on MSFE, however I do not know how to get the MSFE of each model and how to compare it.

When you forecast from an equation in EViews, it tells you the MSE of the forecast. Or, you can use the @rmse function.

ok thx and how can I compute the MSFE within a VAR model?
Because its not given as in Multi or Univariate models....

How are you forecasting from the VAR?

I know how to forecast from the VAR, but I dont know how I can get the MSFE from a forecasted VAR model because its not given as with regular forecast of UNI and MULTIvariate models.

Ok, then use @rmse

i think the question veuve asks above about forecast comparison is more specifically about forecast encompassing. So for example if I have calculated for my series of GDP a univariate, conditional, and VAR model, I want to check whether one forecast incorporates all the relevant information in competing forecasts or if a bivariate combination of forecasts are able to give a more accurate result.

According to Diebold (2007, p.264), one can only gain something from combining forecasts if no forecasts incorporates all the relevant information already, i.e. no forecast is encompassing the other.
Thus before combining the forecasts one has to test whether one forecast is encompassing the other or not?

In other words I regress the 2 forecasts I want to combine on the actual values of GDP. Afterwards I test whether one of the coefficients is equal to 1 and the other equal to 0, in which case the first forecast would encompass the latter, so there shouldn’t be gains from combining the forecasts.

But the question here is how do I do this on EViews?

Any help will be really appreciated?

Veuve's question was a lot more vague than yours, but it is possible that that is what he is asking.

Anyhow, to perform the steps you outline, you pretty much follow the steps you outline. Create the two forecast series (either from an equation or from a VAR (using the VAR forecast add-in, or via a model)), then regress them on the actual values using another equation. Then perform a Wald test of whether a coefficient is equal to 1.

hey gareth,

that is exactly what i did.

But the problem is that every possible combination I regress on GDP, e.g. unidyn - condyn, unistat - constat, unidyn-vardyn, etc., results in either ß1=1 and ß2=0, e.g. unidyn encompasses condyn. Thus, I have no possible gains from a combination. Is that possible or did I do something wrong in my forecast?

I further want to calculate the RMSE from my combined models using the arithmetic average of forecasts and the RMSE from combined models using the weighted arithmetic average. To find the weights, I used the formula from Diebold (2007, p.267). How can I do this on EViews?

I have to add that this is one of the coolest programs ever. Thank you guys for a great job.

I don't know enough about the process you are describing to be able to comment on what you might or might not have done wrong.

However, as long as you have two series, you can easily compute the RMSE using the @rmse function.

Where do I have to type in the @rmse funtion?

this doesnt work: ls gdp constat unistat @rmse