The RATS Software Forum

Hi Tom,
I'm wondering how can i change the title of this topic :maximum likelihood estimators to VAR study?

Also, according this task below:

[quote][/quote]Use the LS estimate and compute point forecasts ^y86(1),^ y86(2) (that is, the
forecast origin is the last quarter of 1968) and the corresponding MSE matrices
^Σy(1),^ Σy(2), ^Σ^y(1), and ^Σ ^y(2). Use these estimates to set up approximate
95% interval forecasts assuming that the process yt is Gaussian.

Where can i get some interesting explanations about how to compute estimated forecast MSE matrices? for h=1 it's easy but as you know the tricky point is for h=2 when i need to compute the omega matrix like in the Lütkepohl's book: New Introduction To Multiple Time Series Analysis p.99?

Thanks Tom

Nabtheberliner wrote:Hi Tom,
I'm wondering how can i change the title of this topic :maximum likelihood estimators to VAR study?

Start a new topic (which I already did). Don't rename an existing one since anyone looking for information on what you were doing before won't find it otherwise.

Nabtheberliner wrote: Also, according this task below:

Use the LS estimate and compute point forecasts ^y86(1),^ y86(2) (that is, the
forecast origin is the last quarter of 1968) and the corresponding MSE matrices
^Σy(1),^ Σy(2), ^Σ^y(1), and ^Σ ^y(2). Use these estimates to set up approximate
95% interval forecasts assuming that the process yt is Gaussian.

Where can i get some interesting explanations about how to compute estimated forecast MSE matrices? for h=1 it's easy but as you know the tricky point is for h=2 when i need to compute the omega matrix like in the Lütkepohl's book: New Introduction To Multiple Time Series Analysis p.99?

Thanks Tom

I would recommend not bothering with that. So-called "exact" calculations of standard errors for IRF's and forecasts was one of Lütkepohl's research interests, and it shows in the book. As you can see, it involves quite a bit of number-crunching. A more common way to compute the distribution of forecasts taking into account uncertainty in the coefficient estimates is with Monte Carlo integration.

Hi Tom,
First question is , i don't see in the interface New Topics? to create one about Impulse responses and Variance decomposition.
Concerning this topic, here my second question. On the wbesite of Estima i looked for the @Bernanke.src procedure but found nothing except bernankemihovqje1998.zip

My last question is about the in and out sample forecasts, in order to have a comparison between the series itself and the forecast.
Thanks a lot Tom for your time
Nabih

Sorry Tom , my last wasn't clear, i'm looking for some documents, i didn't find it in the user's guide.
N.

Nabtheberliner wrote:Hi Tom,
First question is , i don't see in the interface New Topics? to create one about Impulse responses and Variance decomposition.

Did you mean to open a new thread? You just do "NewTopic".

Nabtheberliner wrote: Concerning this topic, here my second question. On the wbesite of Estima i looked for the @Bernanke.src procedure but found nothing except bernankemihovqje1998.zip

The BERNANKE procedure is quite obsolete. It pre-dated the CVMODEL instruction. You should use that instead.

Nabtheberliner wrote: My last question is about the in and out sample forecasts, in order to have a comparison between the series itself and the forecast.

The @UFOREERRORS procedure sounds like what you want.

Hello Tom,
I'm trying to run a full VAR forecasting with graph applying Lutkepohl VAR forecasting progr.P.098 but it doesn't recognize the upper variable and i get an empty range on the series forecurrency(1) and the other two? the currencies file is attached
I don't see the mistake.
Thanks Tom

Sorry the empty range is about the series upper not the forecurrency(1) as you can see

Your data end in 2009:07:07, but you're doing dynamic forecasts starting in 2009:7:9:

FORECAST(MODEL=CURRENCY,FROM=2009:07:09,STEPS=10,RESULTS=FORECURRENCY,STDERRS=STDERRCURRENCY,print)

Because there's no data in 2009:7:8, the forecasts are NA's.

Thank you Tom!
Just a question, why graphically is it so flat? is it normal?

If the data are returns to currency (or to stocks, or to almost any liquid financial asset), yes, the forecasts will be quite flat with a wide error band.

Thanks you confirmed what i thought, wasn't not sure 'cause i'm getting started with econometrics for finance.