Forecasting

[EViews Gareth]

Post your workfile.

[ah09185]

I have 5 countries. for each country i found rt= log(pt)-log(pt-1). I inserted these log returns into eviews for each 5 countries. Thats all I did. Then I tested for ARCH effects, correlation, normality, I estimated the models and then when I do the forecast I get NA in the forecasting variance and covariance and the same out of sample forecasted values. Which file do you want me to upload?

[EViews Gareth]

Your EViews file.

[ah09185]

Hello,
I actually corrected my mistake.Thank you anyway

[j52]

Hey Gareth,

I'm still confused with the forecasting. Can you maybe explain me the difference between the statistic and dynamic forecast? I mean I read the guide but it's still not clear.
The statistic forecast always provides better fc evaluation results especially for the bias proportion. But I'm more interesated in a forecast for a longer period. So is there a possibility to use the statistic fc or do I have to use the dynamic.

And still my question from before; why do I get a longer statistic forecast (for example 6 month ahead) when I estimate the forecast with level data instead of d(x)???

Thanks a lot.

[EViews Gareth]

http://www.eviews.com/Learning/forecasting.html

[j52]

Thanks, but like I said, I've already read everything and still don't quite understand it.

Lets say, I want to have a forcast for 6-month including
-c
-ARMA ar(1) ar(2) ma(1)
-Seasonal dummie variables @expand(@month,@drop(12))
-lagged independent variables x(-6)
-Auto-series d(y) d(x)

With the ARMA terms I have to use dynamic forecast, which makes sense to me. But concidering, not to use the ARMA terms:

-do I get a 6-month forecast with the statistic forecast?

using

-d(y) d(x(-6)) --> no
-d(y) dx(-6) --> no
-dy d(x(-6)) --> yes
-dy dx(-6) --> yes

why is this? And what does it mean?
Regarding you link; they adress the topic with the auto-series. But slide 53; they have the problem that no forecast is being generated using dy instead of, like in my case, a even longer forecast. Would be great if you could explain this to me!

[j52]

maybe to add this:

it also works with
y x(-6)

which also makes sence to me, since I have the data for y(+6) which is x.

So why shouldn't I be able to forecast this with differenced data?

[EViews Gareth]

Just think about what it is that you are forecasting, and then think about whether the data is available for that forecast or not.

If your dependent variable is D(Y) then EViews knows that this is Y - Y(-1). If you want to perform a static forecast of Y, you need values of Y(-1).

[j52]

The data is available for the forecast! I mean that it works using dy d(x(-6), shows that it is possible. But I don't understand why it doesn't work with d(y) d(x(-6)), which I would prefer because I need the forecast for y and not d(y).

[EViews Gareth]

Because Y(-1) isn't available!

[j52]

okay, so using dynamic forecast the d(y) value is calculated by the estimated y, why it doesn't work w/ the statistic forecast!?


But I have still another question using a normal regression like before: d(y) d(x(-6))

- if I calculate the difference using the option d(y) --> I get the same numbers for a dynamic and statistic forecast
- if I calculate the level data using the option y --> I get different results for dynamic and statistic forecast

why?

I would think that especially since the calculation for the growth, the differenced data is the same, shouldn't the results for the absolut data be the same as well!? Or how is it calculated?

Thanks again

[EViews Gareth]

I think you're confused as to the difference between a dynamic and a static forecast.

If you've estimated a model in which Y depends on past values of Y, a static forecast will use the actual values of past Y. A dynamic forecast will use forecasted values of past Y. Of course they will be different.

[j52]

No, I actually understand that part ;) I mean that the first thing you read and not too hard to understand.

And from your answers I get that the d(y) is also treated as dynamic value, since it's been calculated using y(-1), right?

But now I don't see why I get the same results forecasting d(y) and different results for forecasting y.
(not considering diffrences to the lenght of the forecast, just the figures for the in_sample forecast)

Because I thought either the values are treated differently by Eviews (regarding the discussion we had earlier), but then the d(y) should also give different results. Or it's hte same as the forecasted d(y) values suggest, but then I don't understand I ask to give the output as y, why there are different results b/w dynamic and statistic.
I figured the growth numbers would just be added up to get the actual values.

You see what I mean? There is probably just a simple thing I'm missing. But I would like to understand the differences so that I know which one I need to use. So thanks again and hope you can help me.

[adam_economist]

Hi,

Sorry to bring up an old post. I am also trying to understand the below.

if I have a sample from t=1 to t=100, estimate a model: dlog(y) c dlog(x) over t=1 to t=90 and forecast over the remaining ten periods (which data for both y and x are available), I can choose between static and dynamic. If I choose to forecast dlog(y) the is no difference between the static and dynamic forecast. However, if I choose to forecast y there is a difference between the dynamic and static forecast.

I'd like to understand why this is? if you take the forecast of dlog(y) and add this to log(y(-1)) to get the forecast of log(y), and then add the forecast of dlog(y(1)) and so on then take the antilog of each observation to get y in each t.. it is different to what the dynamic forecast produces.

Can you help me understand what exactly is happening here?

Thanks

Adam