mix of dynamic and static forecast

[kiber_master]

Hello!

Could you explain, why in my sample for dynamic forecast EViews use values of dependent variable on the forecast period?

In the workfile y is a serie that have values on the forecast period (2012-2020), y1 is a serie that have no values on the forecast period

[EViews Gareth]

You don't say what period you forecasted over...

But if you forecast from 2000-2030, you'll get forecast values up until 2010, which is the last period for which you have data for your independent variable. Obviously you cannot forecast for periods for which your X variable has no data.

[kiber_master]

the forecast period (2012-2020), it was said

[EViews Gareth]

Sorry, I must be blind. Anyway, my answer remains the same.

[kiber_master]

Your answer is clear, but I can't understand, why I suddenly get the forecast on 2013-2015. For dynamic case and for variable y it shouldn't be at all, because we can't calculate forecast on 2012 (using ar(1), 2011 is missed). But if there is a value at 2012 for y, the forecast is calculated. I think, it isn't a dynamic forecast, because it shouldn't take into account 2012 actual value, only fitted by the model.

If I explained smth not very clear, please, give me a sign!

Thank you

[EViews Gareth]

I don't get a forecast for 2013-2015, so I don't know.

[kiber_master]

So, if you:
1. Open the attached workbook.
2. Open equation tsls.
3. Estimate y c x1 ar(1) (x2 - instrumental, const and lagged vars added) for a period 1980-2011.
4. Click forecast and set forecast dates for 2012-2030.
5. Deselect option "Insert actual values...".
6. Calculate forecast.
the variable yf wil be empty?

[startz]

I do get a forecast for 2013-2015. Since you have X1 from 2011-2015 this is as it should be I think.

[kiber_master]

But there is ar(1). So, the value at 2013 depend on forecasted y at 2012, but it is empty.

[startz]

But there is ar(1). So, the value at 2013 depend on forecasted y at 2012, but it is empty.

I think you're confusing y - yforecast with y-x*beta. The former requires the lagged residual, the latter doesn't. At least I think that's what's going on.

[kiber_master]

There shouldn't be any y for dynamic forecast. Dynamic means using only forecasted values, like yforecasted-x*beta, doesn't it?

[startz]

I don't think this is right. Dynamic forecasting refers to forecasting out the y. The AR(1) comes from forecasting out from the last observed residual.

[kiber_master]

I think it is not correct to use values out of the sample periods we set in equation. What if I don't want to use them for calculation bu check them with forecasted? This behavior is inconsistent, because, if there is no missing values at the end of the sample, to calculate y_forecasted(2013) you use y_forecasted(2012), not y_actual(2012). It isn't understandable for me(

[EViews Gareth]

If you ask EViews to do a dynamic forecast, then EViews will use actual values of Y(-1) for the first forecasted period. It has to do that, other wise you'd never be able to do a dynamic forecast.

I believe what you're questioning is the fact that if Y(-1) isn't available for the first forecasted period, EViews will automatically move the forecast sample up until it does have a valid, actual, value for Y(-1). When EViews does this, and you are doing the forecast interactively, you can spot it from the fact that EViews tells you that it adjusted the forecast sample. If you're doing it in a program, though, there is no way to know, which is unfortunate.

[kiber_master]

Thank you for the explanation! Now all is clear for me!